The Healthcare Revolution That Wasn't.

A Summary of Trends and Challenges #HIMSS24. #AMIA. #AMDIS.

The past two decades have seen massive investments in healthcare information technology, with the promise of transforming and improving healthcare delivery. Yet despite billions spent, the results have fallen far short of expectations. This article examines the 7 major waves of healthcare IT spending since 2000, and why the much-hyped revolution has failed to materialize.?

Wave 1 - Electronic Health Records (EHRs):?

The widespread adoption of electronic health records (EHRs) was kicked off in 2009 with the HITECH Act, which provided financial incentives for eligible providers to implement EHR systems. The goal was to move the healthcare industry away from paper charts and toward digital documentation and information sharing. Adoption of EHRs quickly skyrocketed, with usage rising from less than 10% in 2008 to over 80% by 2015. However, physicians soon found themselves frustrated with poorly designed EHR systems that were cumbersome to use. Entering documentation and orders took more clicks and screens compared to paper charts. EHR vendors filled screens with irrelevant data fields and templates that didn't match clinical workflow. Interoperability between different EHR systems was also lacking, meaning data exchange and coordination of care between providers was difficult.

While EHRs made patient data more accessible digitally, many systems implemented rigid workflows and cluttered interfaces that physicians despised using. Surveys showed high rates of dissatisfaction and even burnout attributed to EHR burdens. This worst-of-both-worlds outcome delivered neither the benefits of digital systems nor the usability of paper records.?

On the financial side, the hoped-for cost savings and efficiency gains from EHR adoption have also been underwhelming so far. Transition costs were high and productivity slowed as physicians spent more time on documentation. The benefits in terms of reduced duplicative testing, administrative overhead, and paper records have not materialized to levels anywhere close to the investment. Overall, the value realized from the digitization of health records using today's EHR systems has fallen well short of the industry's vision. There is still potential for improvement, but only if systems become more physician-centric and focus on clinical efficiency rather than billing maximization.

Wave 2 - Enterprise Data Warehousing:?

In the 2000s, healthcare organizations made big investments in enterprise data warehouses (EDWs) with the goal of aggregating data across various business units and source systems into one place. This was intended to harness the power of "big data" to enable advanced analytics, decision support, and improved operations.???

Healthcare enterprises have numerous disparate systems storing patient data, clinical data, insurance claims data, and more. But each system and department tends to have its own siloed dataset. EDWs promised to integrate all this data into unified repositories that could be analyzed enterprise-wide.? Unfortunately, early EDW efforts faced major challenges:

• Integrating scattered islands of departmental data proved complex and expensive, with constant needs to reformat, map, and validate incoming data.
• Ownership conflicts emerged as departments saw their data being appropriated into the EDW. There was reluctance to share "their" data.
• Disconnected legacy systems made real-time data integration impossible. EDWs were limited to batch updates, sometimes days old.
• Each facility or data domain required its own customized data warehouse designs, preventing an enterprise-wide analytics platform.

Due to these difficulties, much of the data collected into EDWs went underused. The promised revolution in organization-wide analytics and decision making never took hold. The data was there, but effectively accessing and leveraging it proved prohibitively resource-intensive in many cases. Like many waves of healthcare IT investment, EDWs were adopted for the promise of big data rather than clear needs. The hype exceeded the reality.

Wave 3 - Healthcare Information Exchanges (HIEs):

Healthcare information exchanges (HIEs) emerged as a solution to the interoperability challenges between disparate electronic health record (EHR) systems. The goal of HIEs was to allow health data to be shared and exchanged securely across organizations, such as hospitals, clinics, payers, and more.

However, several issues have hindered the effectiveness and adoption of HIEs:

• Competing HIE platforms and lack of standards - Different regions and states adopted different HIE technologies, preventing seamless nationwide data exchange.
• Patient privacy concerns - Patients fear data leakage or misuse, even with HIE safeguards. Consent policies vary across states, limiting participation.
• Inadequate financial incentives - Providers bear the cost of HIE integration with little financial upside, apart from government grants and incentives that were temporary.
• Spotty connectivity - Some regions have robust HIE networks while others have minimal participation and adoption. Coverage is far from universal.
• Data gaps - HIEs typically don't include images, full chart notes, and other clinically useful data that remains trapped in siloed EHRs.

Due to these challenges, HIE adoption has been sluggish and data exchange remains fragmented. Despite federal funding and support, HIEs have not delivered on their promise of seamless health data exchange between medical organizations. Critical information is still not accessible to providers when and where they need it at the point of care. Like other healthcare IT efforts, the hype around HIEs has exceeded the reality thus far.

Wave 4 - Natural Language Processing and Evidence Automation:?

Natural language processing (NLP) and conversational AI tools saw heavy investment in healthcare over the last decade for their potential to automate clinical documentation and workflows.

Products like voice assistants, chatbots, and medical scribes based on NLP were pitched as solutions to physician burnout by reducing the EHR documentation burden. Other NLP algorithms promised to extract insights from unstructured physician notes or patient-reported data.

Despite the potential, NLP and automation have faced challenges getting traction:

• Integration into complex EHR workflows proved difficult and expensive, limiting deployments.
• Error rates for voice transcription and other NLP tasks remained too high for practical clinical use without extensive human verification and editing.
• Clinicians were slow to trust and adopt experimental new technologies like chatbots for fear of disruption, errors, and added work.
• Customization was required for each use case, preventing widescale reuse of tools. NLP performance suffered when applied broadly across specialties.?

Medical scribe applications using NLP demonstrated productivity improvements in narrow documentation use cases. But the promised revolution in automated documentation and workflow support has not arrived. Like other AI technologies, NLP's limitations became apparent once moved from concept to full-scale clinical implementation and adoption. The gap between hype and reality persists.

IBM's Watson is a prime example of the gap between hype and reality when it comes to applying AI in healthcare.? Watson was touted as a breakthrough AI system that could understand natural language, read medical records, diagnose patient conditions, and recommend treatments better than human clinicians. IBM signed high-profile partnerships with healthcare systems and boldly predicted Watson would revolutionize medicine.

The reality fell far short:

• Implementing Watson required enormous effort customizing it to each use case, showing it was not the plug-and-play AI miracle it was made out to be.
• Clinical users found Watson's treatment recommendations often irrelevant or even potentially harmful, without the reasoning to back them up.
• Technical limitations became apparent, like an inability to synthesize patient data from multiple sources into a full clinical picture.
• Adoption languished as clinicians found using Watson overly time-consuming compared to uncertain benefits.

By 2018, some of IBM's earliest healthcare partners ended pilots with Watson without implementation. The backlash was swift - Watson went from hailed as healthcare's AI savior to an example of hype gone wrong.

The failures of Watson in healthcare illustrate the immense gap between marketing hype and true progress in applying advanced AI. Simple demos and proofs-of-concept are far from clinical-grade intelligence. As the Watson case demonstrated clearly, overpromising on immature technology damages trust and adoption of even the legitimate use cases. The way forward is incremental evidence-based clinical integration, not lofty pronouncements detached from reality.

Wave 5 - Clinical Decision Support (CDS)

Clinical decision support (CDS) tools aim to enhance medical decision making by providing insights and recommendations to clinicians at the point of care. CDS encompasses a broad range of health IT applications, from basic alerts and reminders to advanced analytics and AI-driven diagnosis support.

CDS adoption saw a surge in the 2010s driven by federal meaningful use incentives and an expanding set of tools built into electronic health records (EHRs), including:

• Drug-drug interaction alerts that warn of potentially dangerous medication combinations.
• Clinical protocol alerts that ensure best-practice adherence for conditions like sepsis.
• Contextual recommendations that surface patient-specific guidelines based on health factors.
• Predictive algorithms that identify patients at risk for complications or readmission.

These and other CDS capabilities have shown promise in preliminary studies and trials:

• Reducing adverse drug events and costly medical errors.
• Improving adherence to clinical practice guidelines, lowering variability.
• Enabling personalized medicine tailored to the patient's profile.
• Supporting clinical workflow versus interrupting it.

True CDS success depends on intelligent integration into provider workflows, presenting useful insights without overloading users. When done well, CDS has immense potential to improve clinical decision making in ways that augment human expertise. Future evolution of CDS towards more predictive, prescriptive, and AI-powered functionality could be transformative if providers are partners in the process.

Wave 6 - Cloud Migration

The promise of cloud computing was appealing to healthcare organizations - reduced infrastructure costs, scalable storage and computing, and hands-off technical management. However, migrating entrenched legacy systems to the cloud has proven slow and difficult.

Many hospital IT ecosystems consist of decades-old, tightly coupled on-premises systems that can't simply be lifted and shifted to the cloud. Extensive re-architecting is required. Even then, performance and reliability issues emerge requiring re-optimization.

The high costs of cloud migration have also dampened adoption:

• Consultant fees for complex redesign of connectivity, dependencies, and data flow.
• Integrating new cloud capabilities into customized legacy workflows requires custom coding.
• Unanticipated needs to retrain clinicians on updated workflows and interfaces.
• Managing hybrid cloud/on-prem environments during lengthy transitions.
• Redesigning data governance, security, recovery systems for the cloud.

Rather than a turnkey solution, cloud migration has proven to be a lengthy, expensive IT project undertaking. For many hospitals, keeping mission-critical systems like EHRs on-premises has been preferable to the arduous transition. While cloud adoption is inevitable, it requires extensive upfront investment in system overhaul and optimization. This has limited the cloud's transformational impact thus far.

Wave 7 - Artificial Intelligence

Artificial intelligence (AI) stands poised to be the next big wave of digital transformation in healthcare. The hype cycle is in full swing, with predictions of AI revolutionizing everything from medical imaging to population health analysis. But actualizing the great promise of healthcare AI has proven difficult thus far. Under the surface of flashy demos and proofs-of-concepts, deep challenges remain around implementing effective and reliable AI that clinicians can trust and productively use. This article examines the new wave of AI in healthcare, the areas of promise, and the very real challenges of turning cutting-edge algorithms into clinical-grade intelligent tools.

Driving the AI Explosion?

The past decade has seen an explosion of academic research and industry investment in healthcare AI. What factors are suddenly making AI viable for healthcare?

• Data Volume - Electronic health records (EHRs) and medical devices are generating vast troves of structured health data. Combined with natural language processing to tap into unstructured clinical notes and reports, rich datasets are available to train AI models.
• Compute Power - GPU computing has created exponentially more powerful and cost-effective infrastructure for intensive machine learning model development. Cloud platforms provide expandable compute.
• Advanced Algorithms - Techniques like deep neural networks, reinforced learning, and transformers allow more human-like pattern comprehension from complex healthcare data.
• Maturing Infrastructure - Frameworks for model development, monitoring, and deployment help transition algorithms from research into production.

With this perfect storm of health data, compute power, innovating algorithms, and infrastructure, AI looks poised to penetrate healthcare in a big way.?

Consensus on High-Potential Areas for Healthcare AI

• Medical Imaging - Analyzing scans and pathology slides for early disease detection and diagnosis support shows massive promise. AI can spot visual patterns the human eye misses.
• Clinical Decision Support - AI could surface timely, relevant insights to doctors at the point of care based on the patient context. This augments human expertise.
• Population Health - Identifying at-risk patients and guiding interventions across populations could make prevention and care management more proactive and personalized.
• ·?????? Patient Monitoring - AI sensing could provide continuous tracking of vital signs, chronic disease management, and early complication warnings.
• ·?????? Drug Discovery - Machine learning can uncover novel molecules, identify new applications for existing drugs, and vastly accelerate clinical trials.
• Administrative Tasks - Automating paperwork, coding, scheduling, and other clerical tasks could free up clinicians while reducing costs.
• Precision Medicine - Tailoring diagnostics, treatments, and recommendations to each patient's genomic profile and health history is made possible by AI.
• Virtual Health Assistants - AI-powered chatbots and voice interfaces provide conversational advice and care guidance as personalized virtual assistants.

The list of promising applications goes on. With its ability to spot patterns and derive insights from massive, complex datasets, AI is poised to augment human capabilities across nearly every aspect of healthcare delivery.? Yet for all its promise, healthcare AI still faces immense real-world challenges to matching the hype. Turning concepts and isolated successes into enterprise-wide impact has proven difficult. Why is this? Here are some of the biggest impediments:

• Data Complexity - Healthcare data is messy, dispersed, multi-modal, irregularly structured, and growing exponentially. Clean, unified datasets are the exception. This impacts model accuracy.
• Algorithmic Bias - Pattern bias lurks due to uneven, incomplete health data. Correcting for things like gender, age, and racial bias requires tremendous vigilance.
• Regulatory Uncertainty - Strict regulation of software touching patient health slows innovation cycles. Liability for algorithm errors has chilled developers.
• Integration Hurdles - Ingesting and outputting AI insights across convoluted IT systems with bespoke workflows adds cost and limits adoption.
• Clinician Hesitance - Doctors are wary of black-box algorithms guiding care without explaining reasoning and earning trust over time.
• Interoperability Gaps - Data trapped in silos due to lack of standardization severely limits model inputs and visibility.
• Data Privacy Concerns - Patient fears around sharing health data, even for research, slows the data gathering needed for robust models.
• The Fragmented Status Quo - Healthcare's inertia keeps many pilots isolated. Diffusion of successful models at scale is rare, preventing transformation.?

Bridging the Healthcare AI Chasm

How can these hurdles be overcome? Here are some ways forward:

• Increase access to open, de-identified health datasets for research and development. This expands innovation beyond deep-pocketed Big Tech firms.
• Move toward standardized data schemas, code systems, and APIs to ease system integration hurdles.
• Employ techniques like explainable AI and confidence scoring to increase algorithm transparency and trust.
• Incentivize collaboration between developers, providers, and EHR vendors to co-design AI/workflow fusion rather than force-fitting.
• Develop model governance frameworks addressing rigorous validation, monitoring, and human oversight of algorithms.
• Embrace incrementalism - pilot small for clinical feedback before scaling. Avoid overpromising to build provider buy-in.
• Incentivize health systems to share implementation successes using agreed outcome measures.
• Fund studies comparing algorithm performance to legacy decision protocols to prove value.
• Increase investment in rapid-turnaround pilot environments modeling real-world clinical data and systems.

No single solution will unlock the vast promise of healthcare AI. But purposeful innovation, extensive collaboration, and lessons from past overhyped IT efforts can steadily bridge the gap between potential and practice.

Cautious Optimism Warranted, But…

When vision is balanced with pragmatism, there are real grounds for optimism around AI in healthcare. The technology has progressed tremendously in both capability and viability. It holds the potential to augment clinical practice, accelerate discoveries, streamline bureaucracies, and make care more predictive and personalized. Much like the advent of electricity, its applications appear limitless. But as with any new technology, hype naturally exceeds present reality. Fulfilling the promise of AI in healthcare requires extensive collaboration, pilot-driven iteration, and patience as providers, patients, regulators, and partners are brought onboard. With deliberate strategy and execution, AI can gradually become a cornerstone of data-driven, evidence-based 21st century medicine. But a measured approach is required to nurture this budding potential without overpromising. We are at the very early stages of the AI journey. With wise shepherding, AI will gradually transform from radical concept to reliable utility. However, the lessons of past inflated expectations around healthcare IT revolutions must be heeded.

Digital Transformation, Anyone?

The traditional model of healthcare, which is based on episodic, disease-focused, clinic-centric, and clinician-controlled interventions, is being challenged by a new paradigm of healthcare, which is based on continuous, patient-centric, wellness and quality of life focused, anywhere, and patient-empowered interventions. This new paradigm of healthcare is also driven by the availability and analysis of 360-degree, multimodal personal, public, population, physical, and social data, which can provide insights into the health status, needs, preferences, and behaviors of individuals and populations. This document summarizes the main trends and challenges of healthcare transformation in the digital era, and discusses some of the implications and opportunities for healthcare providers, policymakers, and researchers.

The following are some of the key trends that are shaping the healthcare transformation in the digital era:

Personalization: Healthcare is becoming more personalized, as digital technologies enable the collection and analysis of individual data, such as genomic, proteomic, metabolomic, microbiomic, behavioral, environmental, and social data. These data can be used to tailor interventions, such as diagnosis, treatment, prevention, and wellness, to the specific needs, preferences, and characteristics of each patient. Personalization can also enhance patient engagement, empowerment, and satisfaction, as patients can have more control and choice over their own health and care.

Prevention: Healthcare is shifting from a reactive to a proactive approach, as digital technologies enable the detection and prediction of health risks, such as diseases, complications, and adverse events, before they manifest or worsen. Prevention can also involve the promotion of healthy behaviors, such as physical activity, nutrition, sleep, and stress management, through digital platforms, such as mobile apps, wearables, and sensors. Prevention can reduce the burden and cost of healthcare, as well as improve the quality and length of life of patients.

Integration: Healthcare is becoming more integrated, as digital technologies enable the coordination and collaboration of different stakeholders, such as patients, clinicians, caregivers, providers, payers, regulators, and researchers, across the continuum of care, from primary to tertiary, from acute to chronic, and from physical to mental. Integration can also involve the interoperability and sharing of data, information, and knowledge, across different sources, platforms, and systems, such as electronic health records, personal health records, health information exchanges, and cloud computing. Integration can improve the efficiency, effectiveness, and safety of healthcare, as well as the continuity and quality of care of patients.

Innovation: Healthcare is becoming more innovative, as digital technologies enable the creation and adoption of new solutions, such as devices, apps, algorithms, platforms, and systems, that can address the unmet needs, gaps, and challenges of healthcare. Innovation can also involve the application and adaptation of existing solutions, such as artificial intelligence, machine learning, natural language processing, computer vision, robotics, blockchain, and internet of things, to the healthcare domain. Innovation can enhance the accessibility, affordability, and scalability of healthcare, as well as the outcomes and value of care of patients.

Healthcare transformation in the digital era is a complex and dynamic process, that involves multiple trends and challenges, that can have significant implications and opportunities for healthcare stakeholders. Healthcare transformation can offer the potential to improve the health and well-being of individuals and populations, as well as the performance and sustainability of healthcare systems and, therefore, calls for a collaborative and multidisciplinary approach, that can balance the benefits and harms, the opportunities and threats, and the rights and responsibilities, of all stakeholders healthcare in the digital era.

Data liquidity is key for healthcare organizations that are looking to provide insights into the health status, needs, preferences, and behaviors of patients, as well as the performance, efficiency, and quality of care delivery processes. However, data alone is not enough to achieve these benefits. Healthcare organizations need to transform their data assets into data insights, which are actionable, relevant, and timely information that can support decision making and improve outcomes. Data-driven insights can help healthcare organizations achieve a number of benefits, such as:

• Healthy patients: Data-driven insights can enable healthcare organizations to provide personalized, preventive, and proactive care to patients, based on their individual characteristics, risk factors, and needs. This can improve patient outcomes, reduce readmissions, and enhance patient satisfaction and loyalty.
• Low healthcare cost: Data-driven insights can help healthcare organizations optimize their resource allocation, reduce waste and inefficiency, and identify opportunities for cost savings and revenue generation. This can improve their financial performance, sustainability, and competitiveness.
• More visibility into performance: Data-driven insights can help healthcare organizations monitor and evaluate their clinical and operational processes, identify gaps and areas for improvement, and implement best practices and evidence-based interventions. This can improve their quality of care, safety, and compliance.
• High staff and consumer satisfaction rates: Data-driven insights can help healthcare organizations engage and empower their staff and consumers, by providing them with feedback, recognition, and incentives, as well as access to information, education, and support. This can improve their motivation, productivity, and retention.

To achieve data-driven insights, healthcare organizations need to follow a systematic process that involves four steps:

• Curation: Healthcare organizations need to collect data from various sources, such as electronic health records, medical devices, sensors, surveys, social media, and external databases. They need to ensure that the data is accurate, complete, and consistent, and that it complies with ethical and legal standards.
• Analysis: Healthcare organizations need to analyze the data using appropriate methods, such as descriptive, predictive, and prescriptive analytics, to generate insights that answer specific questions, address specific problems, or support specific goals. They need to use advanced tools and techniques, such as artificial intelligence, machine learning, and natural language processing, to handle the complexity and volume of the data.
• Visualization: Healthcare organizations need to present the data insights in a clear, concise, and compelling way, using charts, graphs, dashboards, and stories, to communicate the key messages and recommendations to the relevant stakeholders. They need to use interactive and dynamic features, such as filters, drill-downs, and alerts, to enable the stakeholders to explore the data and gain deeper insights.
• Implementation: Healthcare organizations need to act on the data insights, by implementing changes, interventions, or solutions that can improve their clinical and operational outcomes. They need to monitor and measure the impact of their actions, and adjust them as needed, based on the feedback and results.

?

Data-driven insights are essential for healthcare organizations that want to achieve healthy patients, low healthcare cost, more visibility into performance, and high staff and consumer satisfaction rates. To achieve data-driven insights, healthcare organizations need to collect, analyze, visualize, and act on their data assets, using a systematic and rigorous process. By doing so, they can leverage their data assets to improve their decision making and outcomes. The future beckons.

?

Sanjay M. Udoshi MD | Senior Principal & CMIO

O: 608.661.7603? M: 570.472.7078

[email protected]

Symphony Corporation | Madison, WI | A SEI-CMMI Level 4 Company